Time Series Satellite Data to Identify Indicator of Ecosystem Health

advertisement
This file was created by scanning the printed publication.
Errors identified by the software have been corrected;
however, some errors may remain.
Time Series Satellite Data to Identify
Vegetation Response to Stress as an
Indicator of Ecosystem Health
Judith Lancaster
David Mouat
Robert Kuehl
Walter Whitford
David Rapport
vegetation. Studies of shrub and forest communities (Law
and Waring 1995; Yoder and Waring 1994) demonstrate
that vegetation operating under severe environmental constraints caused, for example, by prolonged drought, will
show lowered photosynthetic rate and therefore decreased
productivity throughout the year. This might be an indicator
of ecosystem health.
Based upon assumptions that numerous stressors affect
arid ecosystems, including climate, grazing, herbicide use
and recreation, and that vegetation composition and cover is
a response to ecosystem stress, it is hypothesized that
satellite data may be used to evaluate ecosystem response
through the use of a vegetation index such as the Normalized
Difference Vegetation Index (NDVI). Ecosystems respond in
different ways to stressors, such as drought, depending on
their structural and functional integrity. Vegetation composition and cover may be seen as a response to stress, either
natural as in the case of climate, or anthropogenic, in origin.
Vegetation responds to water stress caused by severe or
prolonged drought by closing leaf stomata, which decreases
conductance of carbon dioxide from the atmosphere into leaf
tissues and limits the plant's ability to absorb carbon through
photosynthesis (Schulze and Hall 1982).
The research objective specific to the study reported here
was to statistically evaluate the suitability of a suite of
metrics derived from NDVI temporal profiles for discriminating variation in response to climate between mesquite
and grassland systems.
Abstract--One measure of health in an ecosystem is the response of
that ecosystem to an environmental perturbation (such as rain) over
time. Healthy and unhealthy systems may have different phenological response patterns. Grassland and shrubland sites were selected,
and metrics derived from temporal profiles of vegetation index
values from Advanced Very High Resolution Radiometer satellite
data from 1987 through 1993. The temporal profiles show that the
vegetation types may be statistically discriminated, and metric
values demonstrate different responses to rainfall.
The term "ecosystem health" is widely used by scientists,
land owners and managers, and policymakers at all levels to
indicate a condition that is both aesthetically and economically acceptable. However, the expression has varying meaning for different interest groups, and the criteria by which it
is assessed often vary from one ecosystem to another. Before
health can be judged, for any ecosystem, it is necessary to
identify indicators such as keystone species, and acceptable
ranges for measured values of these indicators (Haskell and
others 1992). Haskell and others (1992) argue that each
ecosystem has a specific suite of indicators, however, we
propose that certain ecosystem structural and functional
characteristics may be used as measurements of health.
Net primary production has been identified as the most
important carbon cycle variable for quantifying and comparing biological activity across regions and biomes (Running
1990), and is an indicator of ecosystem function. Changes in
ecosystem structure, such as biomass or leaf area index, may
be identified and monitored using spectral radiance data
acquired by sensors on-board satellites or aircraft to determine species composition and vegetation patterns (Mouat
1995). The repeated measurement of an ecosystem variable
from the synoptic perspective of an aircraft or satellite
sensor permits ecosystem analyses at varying spatial and
temporal scales, facilitating the analysis of ecosystem
dynamics which are reflected in temporal changes in
Methods ---------------------------------An experimental design was developed which involved the
selection of sites in both grassland and mesquite dominated
areas, and verification of their vegetation and soils homogeneity using spectral response of vegetation during periods of
maximum photosynthetic activity. Values for NDVI for each
site at approximately biweekly periods during the growing
season were extracted from satellite imagery, and plotted as
annual temporal profiles for a seven year period from 1987
through 1993. The profiles were smoothed using a locally
weighted regression, and metrics characteristic of phenological response calculated for each site for each year of the
study. Values for the metrics were correlated with seasonal
and annual rainfall totals, and the differences in photosynthetic response between vegetation types statistically
validated.
In: Barrow, Jerry R.; McArthur, E. Durant; Sosebee, Ronald E.; Tausch,
Robin J., comps. 1996. Proceedings: shrubland ecosystem dynamics in a
changing environment; 1995 May 23-25; Las Cruces, NM. Gen. Tech. Rep.
INT-GTR-338. Ogden, UT: U.S. Department of Agriculture, Forest Service,
Intermountain Research Station.
Judith Lancaster is with the Desert Research Institute, Reno, NV. David
Mouat is with the Environmental Protection Agency, Corvallis, OR. Robert
Kuehl is with the University of Arizona, Tucson, AZ. Walter Whitford is with
the Environmental Protection Agency, Las Vegas, NV. David Rapport is with
the University of Guelph, Guelph, Ontario.
255
Satellite Data
maintained climatic and land-use records since 1915, which
is critical for a historical perspective on climatic variability.
Covering an area of over 40,000 hectares, the JER is within
the Chihuahuan Desert, ranges in elevation from 1,260 to
2,833 m, and has a mean annual precipitation of 247 mm
(Agricultural Research Service 1994). Soils contain little
organic matter, and root and water penetration are limited
by a thick caliche layer often over 2 m in thickness (Agricultural Research Service 1994).
According to the work of Buffington and Herbel (1965), the
western portion of the JER originated as perennial grassland, although it is now a spatially diverse mixture of
mesquite dominated shrubland, grassland, and combinations of both vegetation types. This area met the site selection criteria for the project, and four sites were located in
mesquite dominated areas, and another four in grassland
areas in the western portion of the JER (fig. 1), thus providing a replication of study sites for purposes of statistical
analysis.
The Advanced Very High Resolution Radiometer (AVHRR)
is carried on-board the NOAA series of satellites, orbiting
the earth twice daily on a sun synchronous schedule (NOAA
1986). Ground based radiance is collected in five spectral
bands: red (Channell, 0.58-0.68 um), near infrared (Channel2, 0.72-1.10 um), mid infrared (Channel 3, 3.55-3.93 um)
and two thermal infrared bands (Channel 4, 10.3-11.3 urn
and Channel 5, 11.5-12.5 urn) from pixels measuring 1.1 x 1.1 km
in size when the satellite is at a nadir viewing angle. With
heterogeneous land cover types, the 1.1 km pixel size results
in the integration of varying spectral responses, which
makes it suitable for regional scale studies. At these scales,
the AVHRR has been used to monitor herbaceous cover in
Botswana (Prince and Tucker 1986), correlate vegetation
biomass with rainfall patterns in the Sahara (Malo and
Nicholson 1990) and assess biological diversity for California (Walker and others 1992). Mouat (1995) hypothesizes
that the spatial scale of AVHRR is directly applicable to the
examination of ecosystem health, based on the premise that
it is appropriate for the ecosystem processes involved
(Malingreau and Belward 1992).
These studies typically use the Normalized Difference
Vegetation Index (NDVI):
Evaluation of Site Homogeneity
In order to compensate for satellite drift of up to one pixel
in any direction, it was necessary for study sites to be
spectrally homogeneous over a 3 x 3 km area. This homogeneity was established from spatial variance of reflectance in
red and near infrared wavebands using imagery from the
Landsat Multispectral Scanner (MSS) with a pixel size of
80 x 80 m. MSS images for April and August were chosen to
capture the periods of maximum photosynthetic activity for
mesquite and grassland communities respectively. The area
of the eight study sites and four less homogeneous sites (as
controls) was clipped from each image, and red and near
infrared waveband reflectance index values were calculated
for aggregates of 81 MSS pixels in 9 x 9 pixel matrices to
approximate 1.1 km AVHRR pixels. The among pixel variance for the vegetation index values for the 16 aggregated
pixels provided a measure of site homogeneity. An a priori
coefficient of variation of 10% was chosen as the minimum
value for site homogeneity, and study sites met this criteria.
The Department of Geography at New Mexico State University was able to provide AVHRR High Resolution Picture
Transmission (HRPT) images for the JER, covering the
period 1987 through 1993. Preprocessing of the AVHRR
data was carried out by New Mexico State University, and
extraction of the NDVI values for the study sites was done at
the Desert Research Institute. Values for AVHRR red and
near infrared reflectance were calculated from cloud-free
imagery acquired at 10 to 15 day intervals throughout the
annual active photosynthetic cycle and used to calculate
NDVI. An area slightly larger than each 3 x3 km study site
was clipped from each AVHRR image, the NDVI value for
the center pixel of the matrix was extracted, and the resulting values were plotted against time to give a temporal
profile of NDVI response. Some early spring and late summer values for NDVI were higher than expected (the "terminator effect", Goward and Peters, personal communication,
1994). As a result these data were excluded from the study,
resulting in nine to 12 data points between April 29 and
September 19 for each study year.
NDVI = (NIR - RED) / (NIR + RED)
where NIR is reflectance in near infrared wavelengths and
RED is red waveband reflectance. The NDVI minimizes the
effects of topography and atmosphere (Holben and Justice
1981), requires no prior knowledge of ground conditions, and
is sensitive to the amount of photosynthetically active vegetation present (Myneni and others 1992; Tucker 1979).
With synchronous data collected on a daily basis, the
AVHRR offers the most appropriate data set for temporal
studies, and cloud-free images are usually available at 10 to
15 day intervals for the western U.S. Temporal profiles of
NDVI or other indices capture and quantify differences in
the extent and intensity of physiological activity throughout
the growing season. Vegetation temporal response has been
used to detect the effects of short-term drought in New
Mexico, based on the differences between wet (1988) and dry
(1989) years (Peters and others 1993), and to observe phenological differences between natural and cultivated vegetation throughout North America (Goward and others 1985).
Seasonal range conditions in Senegal were monitored using
NDVI time-series data (Tappan and others 1992), and Malo
and Nicholson (1990) found that spatial and temporal patterns of NDVI closely followed rainfall in west Africa.
Site Selection Criteria
A prerequisite for the selection of a study area was current
vegetation with variation in composition and cover which
had originated as a uniform vegetation type. The Jornada
Experimental Range (JER), near Las Cruces, New Mexico,
comprises a mixture ofvegetation types, having experienced
an increase in shrub species and decrease in area occupied
by perennial grasslands since the first vegetation survey in
1858 (Buffington and Herbel 1965). In addition, the JER has
256
New Mexico
........-
o
2.5
'
Scale In kilometers
•
Figure 1-Location of study sites (squares)
and rain gauges (named locales) at Jornada
Experimental Range. Dark gray sites are
mesquite dominated, pale gray sites are grassland areas.
Alulllllrgedibovi
computed for time series, however, a locally weighted linear
or quadratic regression is used for the smoothing rather
than a simple average.
The LOESS smoothed fit illustrated for one o( the grassland sites in 1989 (fig. 2) includes the twice standard error
band around the smooth fit which corresponds roughly to
95% confidence limits for the nonparametric smoothing
(Hastie and Tibshirani 1990). NDVI values for each observation over the 1989 growing season were aggregated for the
four grassland and four mesquite sites respectively, showing
significant variations in phenological response between the
two vegetation types (figs. 3 and 4).
A number ofmetrics derived from NDVI temporal profiles
have been proposed as indicators of ecosystem identification, behavior and response to perturbation, (Samson 1993;
Reed and others 1994). These derived metrics include total
NDVI response measured as an integration of the NDVI
profile, seasonally integrated NDVI, duration of NDVI response, maximum NDVI, date of maximum NDVI, rate of
NDVI increase during the greenup periods, and rate of
NDVI decrease during the brown-off periods.
Six primary metrics calculated from the-yearly profiles at
each of the eight sites are illustrated in figure 5. They are
total NDVI response, or integrated area under the profile
NDVI Analysis
A daily profile of NDVI response or trend over time at a
given site was felt to characterize the distinct pattern or
shape of the NDVI response on the site. Various metrics
could then be derived from the NDVI time profile with the
potential to distinguish the patterns or shapes ofthe profiles
between the two vegetation types. Since NDVI data were
available for only 9-12 dates at each site, a daily profile of
NDVI response for each site had to be estimated from these
9-12 NDVI observations.
An implementation of robust locally weighted regression
or LOESS (Cleveland 1979; Cleveland and Devlin 1988) was
used to estimate the NDVI profile for each ofthe sites in each
of the years. LOESS is a nonparametric function to describe
the relationship between two variables.
In the case of the NDVI profiles the NDVI response was
assumed to be some unknown function ("g") of time:
g(Ti ) or NDVI j = g(Ti ) + ej
where i = 1,2, t and e(i) is random error, and T is time. The
NDVI responses were smoothed as a function of time in a
moving fashion analogous to how a moving average is
257
0.12
0.15
0.10
is
>
is
>
0.10
Z
0.08
Z
0.06
0.05
0.04
0.0
0.02
150
200
250
150
Julian Date
Figure 2-NOVI temporal profile LOESS fit for a
grassland site with NOVI values plotted as points, and
twice standard error limits shown as a dotted line.
with NDVI above 0 ..03; duration ofNDVI response, or number of days with NDVI above 0.03 between April 29 and
September 19; NDVI response between April 29 and June 15;
NDVI response between August 1 and September 19, maximum NDVI; and date of maximum NDVI. Examples ofNDVI
metrics, also given for 1989, are shown in table 1.
The climate data used for the analysis comprised previous
year total rainfall, previous winter rainfall (October-Janu~lrY), current year spring rainfall (February-April), current
year (May and June) rainfall, current year summer rainfall
(July-September), current year total rainfall, and degree
days above 5 °C between January 1 and May 15 for the
current year. The rainfall data were collected at separate
rain gauges for the mesquite sites and the grassland sites
resulting in distinct rainfall values for the two types of sites.
The degree days were common for the two vegetation types.
0.08
Z
0.06
0.04
0.02
150
250
Figure 4-LOESS smoothed fit for aggregated site
NOVI data for mesquite in 1989.
0.10
is
>
200
Julian Date
200
250
Julian Date
Figure 3-LOESS smoothed fit for aggregated site
NOVI data for grasslands in 1989.
Maximum NOVI
0.15
=.13, date of maximum NOVI =248
0.10
is
>
z
0.05
f---~
Total NOVI
Early NOVI
--?
f-
Total NOVI
~
Late NOVI
~
0.0
150
200
Julian Date
Figure 5-LOESS smoothed NOVI profile showing criteria
for deriving the metrics.
258
250
Table 1-Values for NOVI metrics for 1989.
Site1
Early2
1
6
7
8
2
3
4
5
3.459
2.893
3.263
2.954
5.072
6.038
5.934
5.037
Late3
7.358
9.682
10.328
10.459
8.341
9.485
9.066
8.848
Maximum
NDVI4
0.081
0.130
0.137
0.149
0.128
0.150
0.141
0.118
Table 2-Rainfall covariates for which the NOVI metries had a
statistically significant positive (increased) response to
increased rainfall (indicated with u+").
Maximum
dateS
Duration'
132
248
247
245
140
139
135
138
121
110
117
121
153
153
153
153
Previous
year
Grassland
Total NOVI
NOVI duration
Spring NOVI
Summer NOVI
Maximum NOVI
Oate of max NOVI
lSites 1, 6, 7, and 8 are grassland and sites 2, 3, 4, and 5 are mesquite
dominated.
2The "early" variable is the integrated NOVI profiles from April 29 to June 15.
3The "late" variable is the integrated NOVI profiles from August 1 to
September 19.
4Value of maximum NOVI.
sJulian date of maximum NOVI.
6Number of days in which NOVI ~ 0.03.
Mesquite
Total NOVI
NOVI duration
Spring NOVI
Summer NOVI
Maximum NOVI
Oate of max NOVI
The generalized linear model (McCullagh and NeIder
1989) was used to estimate the relationships between the
NDVI metrics and the climate covariates in order to determine whether the NDVI metrics for mesquite and grassland
systems had differential responses to climate. The generalized linear models allow for a broad range of statistical error
model families in the parameter estimation routine, and as
such are not limited to the normal distribution error model.
The response variables for the generalized linear model
were the NDVI metrics. The independent variables or
covariates included the study design variables of vegetation
type (grassland or mesquite) and replicate sites within each
vegetation type as well as the climate covariates. The rainfall covariates were included in the model with separate
regressions for each of the vegetation types since there were
separate rainfall measures for each vegetation type. Degree
days was entered as a common regression for both vegetation types since the degree days covariate was common to
both types. The analysis for regression ofthe NDVI metric on
rainfall covariates was conducted to determine whether the
regression coefficients were significantly different for the
two vegetation types or of similar value for the two vegetation types.
Results
Rainfall
Previous
FebApril
winter
MayJune
+
+
+
+
+
+
+
JulySept
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
to spring rainfall, and the same NDVI metrics for grassland
sites which were predominantly responsive to summer rainfall. The NDVI metrics for the two vegetation types responded
similarly to total rainfall in the previous year. Total, early,
and late NDVI response as well as the maximum NDVI
metrics all increased with previous year rainfall.
Total NDVI response for grassland sites increased with
summer rainfall, whereas the total NDVI response for mesquite sites increased with spring rainfall.
The duration of NDVI response increased with spring
rain for both grassland and mesquite sites. This was probably a function of prolonging the growing season. However,
the duration ofNDVI for grassland sites also increased with
previous winter and summer rain, while the duration of
NDVI response for mesquite sites was significantly related
only to spring rainfall.
Early NDVI response (April 29 to June 15) for mesquite
sites increased with previous year and spring rainfall, while
the early NDVI response for grassland sites was significantly related only to previous year rain. This may have been
a result of the response of a minor C g shrub, grass and forb
component.
Late NDVI response (August 1 to September 19) for
grassland sites increased with May-June and summer rainfall, whereas the late NDVI response for mesquite increased
only with spring rainfall.
The maximum NDVI for sites of both vegetation types
increased significantly with spring and previous year rainfall. As might be expected, the date of maximum NDVI was
significantly later with increased summer rainfall for the
grassland sites, but for mesquite sites the date was significantly later with increased previous winter rainfall as well
as increased summer rain.
Summarizing the differences in the response of NDVI
metrics to rainfall variables it was found that total NDVI
response for grasslands increased with summer rain, whereas
the total NDVI response for mesquite increased with spring
------------------------------------
The results of the best fitting models for each of the
vegetation types are shown in table 2, indicating those
covariates to which the NDVI metrics had a statistically
significant positive response (for example, increased). In no
instance was the "degree days" variable significantly related
to any of the NDVI metrics.
In general the patterns of positive responses of the two
vegetation types to rainfall have some similarities and some
differences. This study was in part designed to identify those
NDVI metrics, if any, which show differential responses for
grassland and mesquite sites to climatic variables as an
indicator of relative health. The primary difference between
the two vegetation types was manifested by NDVI metrics
for the mesquite sites which were predominantly responsive
259
rain. The duration of NDVI response was prolonged in
grasslands by increased previous winter, spring, and summer rainfall, whereas it was prolonged only by spring rainfall in the mesquite. The early NDVI response of grasslands
increased only with previous year rainfall, but the early
response of mesquite also increased with spring rainfall. The
late NDVI response of grasslands increased with May-June
and summer rainfall whereas the late response of mesquite
increased only with spring rainfall. The date of maximum
NDVI for grasslands occurred later only with increased
summer rainfall, but the maximum NDVI for mesquite also
occurred later with increased previous winter rainfall. Maximum NDVI relationships to rainfall were similar for the two
vegetation types since both increased with previous year
and spring rainfall.
Although the information in this document has been
funded wholly or in part by the U.S. Environmental Protection Agency under CR-819549-01 to the Desert Research
Institute it does not necessarily reflect the views of the
Agency and no official endorsement should be inferred.
References
--------------------------------
Agricultural Research Service. 1994. The Jornada Experimental
Range, Las Cruces, New Mexico. Agricultural Research Service,
Las Cruces, New Mexico.
Buffington, L.C. and C.H. Herbel. 1965. Vegetation changes on a
semidesert grassland range from 1958 to 1963. Ecological Monographs. 35: 139-164.
Cleveland, W.S. 1979. Robust locally-weighted regression and
smoothing scatterplots. Journal ofthe American Statistical Association. 74:829-836.
Cleveland, W.S. and S.J. Devlin. 1988. Locally-weighted regression:
an approach to regression analysis by local fitting. Journal of the
American Statistical Association. 83:596-610.
Donovan, LA. and J.R. Ehleringer. 1994. Water stress and use of
summer precipitation in a Great Basin shrub community. Functional Ecology. 8:289-297.
Goward, S.G., C.J. Tucker and D.G. Dye. 1985. North American
vegetation patterns observed with the NOAA-7 advanced very
high resolution radiometer. Vegetatio. 64:3-14.
Haskell, B.D., B.G. Norton and R Costanza. 1992. What is ecosystem health and why should we worry about it? In: Ecosystem
Health, R Costanza, B.G. Norton and B.D. Haskell (eds), Island
Press, Washington, DC. pp. 3-19.
Hastie, T.J. and RJ. Tibshirani. 1990. Generalized Additive Models. Chapman and Hall, London.
Holben, B.N. and C.O. Justice. 1981. An examination of spectral
band rationing to reduce the topographic effect on remotely
sensed data. International Journal of Remote Sensing. 2:115-121.
Law, B.E. and RH. Waring. 1995. Combining remote sensing and
climatic data to estimate net primary production across Oregon.
Ecological Applications. 4(2):272-279.
Malingreau, J.P. and AS. Belward. 1992. Scale considerations in
vegetation monitoring using AVHRR data. International Journal
of Remote Sensing. 13(12):2289-2307.
Malo, AR and S.E. Nicholson. 1990. A study of rainfall and vegetation dynamics in the African Sahel using normalized difference
vegetation index. Journal of Arid Environments. 19:1-24.
McCullagh, P. and JA. NeIder. 1989. Generalized Linear Models.
Chapman and Hall, London.
Mouat, D.A 1995. Remote sensing and ecosystem health: an evaluation of time-series AVHRR NDVI data. In: Evaluating and
Monitoring the Health of Large-scale Ecosystems. D.J. Rapport,
C.L. Gaudet and P. Calow (eds), Springer-Verlag, Berlin. NATO
ASI Series, Vol. 128. pp. 431-446.
Myneni, RB., B.D. Ganapol and G. Asrar. 1992. Remote sensing of
vegetation canopy photosynthetic and stomatal conductance efficiencies. Remote Sensing of Environment. 42:217-238.
Neilson, R.P. 1986. High-resolution climatic analysis and Southwest biogeography. Science. 232:27-34.
NOAA. 1986. NOAA Polar orbiter user's guide. National Oceanic
and Atmospheric Administration, National Environmental Satellite, Data and Information Service, National Climatic Data
Center, Satellite Data Services Division, Washington, DC.
Peters, AJ., B.C. Reed, M.D. Eve and K.M. Havstad. 1993. Satellite
assessment of drought impact on native plant communities of
southeastern New Mexico, U.S.A Journal ofArid Environments.
24:305-319.
Prince, S.D. and C.J. Tucker. 1986. Satellite remote sensing
of rangelands in Botswana II: NOAA AVHRR and herbaceous
vegetation. International Journal ofRemote Sensing. 7:1555-1570.
Reed, B.C., J.F. Brown, D. VanderZee, T.R Loveland, J.W.
Merchant and D.O. Ohlen. 1994. Measuring phenological variability from satellite imagery. Journal of Vegetation Science.
5:703-714.
Conclusions -----------------------------Of primary significance in the results is the demonstrated
statistical discrimination, in a site replicated study, between the two vegetation types with metrics derived from
their NDVI response. Equally important, the results are
consistent with differences in biological responses of the two
vegetation types to rainfall patterns found in other studies
(Donovan and Ehleringer 1994; Neilson 1986; Van Devender
in press).
The effective statistical discrimination of two distinct
vegetation types with satellite derived metrics provides the
potential framework for a technique useful to discriminating these ecosystem characteristics in other settings.
NDVI metrics derived from temporal profiles are a measure of phenological response, and may indicate primary
productivity. Changes in values for individual metrics, or
the suite of metrics, occurring over a period of several years,
might indicate changes in ecosystem structure, and may be
used to monitor the condition, or health, of the system.
We have hypothesized that one measure of health in an
ecosystem is the response of that ecosystem to an environmental perturbation (such as rainfall) over time. A healthier
system may have a different phenological response pattern
than the less healthy system and will maintain this pattern
over time. The less healthy system will have a phenological
response pattern which will diverge from the healthier one.
The research reported illustrates a landscape which was
predominantly grassland in the early 19th century. Following human disturbance, much of the grassland was replaced
by shrubs (dominantly mesquite). We feel that the shrubdominated areas took advantage of winter rains and "green
up" earlier in the year. The AVHRR NDVI temporal response profiles support this observation. Continuing analysis will help to elucidate these relationships and either
support or reject the hypotheses.
Acknowledgments _ _ _ _ _ __
We are indebted to AI Peters, Julie Eggerton, Barbara
Nolen and Marlen Eve (New Mexico State University), Jeff
Herrick (JER), Billy Lassetter, Tim Minor, Tim Wade and
Jody Hatzell (Desert Research Institute) and Haiyan Cui
(University of Arizona) for their contribution to the success
of this project.
260
Running, S.W. 1990. Estimating terrestrial primary productivity by
combining remote sensing and ecosystem simulation. In: Remote
Sensing and Biosphere Functioning, RJ. Hobbs and H.A Mooney
(eds), Springer-Verlag, New York. pp 65-86.
Samson, S.A 1993. Two indices to characterize temporal patterns
in the spectral response of vegetation. Photogrammetric Engineering and Remote Sensing. 59(4):511-517.
Schulze, E.D. and AE. Hall. 1982. Stomatal responses, water loss
and CO2 assimilation rates of plants in contrasting environments. In: Physiological Plant Ecology II: Water Relations and
Carbon Assimilation, Encyclopedia of Plant Physiology, Volume
12B. O.L. Lange, P.S. Nobel, C.B. Osmond and H. Siegler (eds),
Springer-Verlag, Berlin. pp. 181-230.
Tappan, G.G., D.J. Tyler, M.E. Wehde and D.G. Moore. 1992.
Monitoring rangeland dynamics in Senegal with Advanced Very
High Resolution Radiometer data. Geocarto International. 7(1):
87-98.
Tucker, C.J. 1979. Red and photographic infrared linear combinations for monitoring vegetation. Remote Sensing ofEnvironment.
8:127-150.
Van Devender, T.R in press. Desert grassland, mixed shrub savanna, shrub steppe or semidesert scrub?: The dilemma of coexisting life forms. In: The Desert Grassland, M.P. McClaren and
T.R Van Devender (eds), University of Arizona Press, Tucson.
Walter, RE., D.M. Stoms, J.E. Estes and K.D. Cayocca. 1992.
Relationships between biological diversity and multi-temporal
vegetation index in California. Proceedings of American Society
for Photogrammetric Engineering and Remote Sensing. American Congress of Surveying and Mapping, Washington, March
1992. Falls Church, Virginia. American Society of Photogrammetry and Remote Sensing. pp. 562-571.
Yoder, B.J. and RH. Waring. 1994. The normalized difference
vegetation index of small Douglas-fir canopies with varying
chlorophyll concentration. Remote Sensing of Environment. 49:
81-91.
261
Download